High Speed and High Reliability InP/InGaAs Avalanche Photodiode for Optical Communications

Invited Paper High Speed and High Reliability InP/InGaAs Avalanche Photodiode for Optical Communications Kyung-Sook Hyun *, Youngmi Paek, Yong-Hwan Kwon a), Ilgu Yun b) and El-Hang Lee c) School of Electronics and Information Engineering, Sejong University, Seoul 143-747, Korea a) Telecom. Basic Research Lab. ETRI 161 Gajung-dong, Yusong-gu, Taejon, 305-600, Korea b) Department of Electrical and Electronic Engineering Yonsei University, Seoul 120-749, Korea c) School of Information and Communication Engineering Inha University, Incheon City, 402-751, Korea ABSTRACT We present a review of the characteristics of several different types of high speed InGaAs/InP avalanche photodiode (APD)s that we have developed for different guard ring depth and for different main p-n junction shape. The APD structure that we propose consists of a greatly reduced width in InP multiplication layer and a high doping concentrated electric field buffer layer, where we also adopted a floating guard ring and a shaped main junction with recess etching for a reliable operation of an APD. We obtained high reliability APDs, which are tested for two-dimensional gain behavior and for accelerated life tests by monitoring dark current and breakdown voltage. The gain and bandwidth product of the best of our APDs was measured as high as 80 GHz. Keywords: avalanche photodiode, photodetector, reliability, optical communication, III-V semiconductors. I. INTRODUCTION The use of avalanche photodiodes (APDs) in 10 Gbps systems is promising to satisfy the increasing demand of high performance optical transmission systems. However, there exist several problems to be resolved for their use as a high-speed optical detector, including low reliability and narrow structural margins for very high-speed response. Many researches have been focused on the improvement of their performances via techniques such as band gap engineering and optimization of device structures in III-V compound semiconductor. Various APD structures have been developed such as InP/InGaAs separated absorption, grading, charge, and multiplication (SAGCM) structure 1, δ-doped SAGM structure 2, and InAlAs/InGaAs super-lattice structure 3, SACM structure adopting quantum-dot resonant-cavity 4, and floating guard ring (FGR) structure 5,6. Although these structures have offered large gain-bandwidth product and high performances at 1.3 and 1.55 µm wavelength, their merits are greatly affected by the variations of the fabrication parameters. Hyun et al. studied on the breakdown characteristics of InP/InGaAs APD with p-i-n multiplication layer, and Park et al. calculated the effective thickness of a multiplication layer width in APD 7,8. Yuan et al. reported on impact ionization characteristics of III-V semiconductors for a wide range of thickness in the multiplication region 9. In this paper, we report, in the form of a review, on the performance characteristics of several different types of InGaAs/InP avalanche photodiode (APD)s, which are made to differ in guard ring depth and the shape of the main p-junction. The fabrication parameters were varied to observe the electrical and optical characteristics in APD. As mentioned above, the most primary concern in the design and operation of an APD is high-speed *Corresponding Author: E-mail: kshyun@sejong.ac.kr Tel:+82 2 3408 3792, Fax:+82 2 3408 3666 130 Quantum Sensing: Evolution and Revolution from Past to Future, Manijeh Razeghi, Gail J. Brown, Editors, Proceedings of SPIE Vol. 4999 (2003) 2003 SPIE 0277-786X/03/$15.00

performance and long-term reliability to be used as an optical receiver. We therefore measured for this purpose the two dimensional gain characteristics and conducted the accelerated life test by monitoring the dark current and the breakdown voltage. We review a range of issues concerning the optimal design of a high speed APD by varying the very thin multiplication layer. Discussions include the typical characteristics on dark and photo current-voltage, two-dimensional gain profile and frequency responses of APD with one floating guard ring and its simple fabrication method regarding single diffusion process. II. EXPERIMENTS AND DISCUSSIONS 1. APD Structure Fig. 1 shows a schematic diagram of our proposed APD, which is based on SAGCM (separated absorption, grading, charge and multiplication) structure. A schematic cross-sectional view of the InP/InGaAs APDs with recess etching is also shown in Fig.1. p-metal P+-InP (Zn diffused) u-inp n-inp u-ingaasp u-ingaas n-inp (buffer) n-inp (Substrate) SiN x SiN x n-metal Figure 1. Schematic diagram of SAGCM APD with one floating guard ring The epitaxial structure is grown by the metal organic chemical vapor deposition (MOCVD) method. The u- InP layer was designed to be 3.5 µm thick and the thickness of the absorption layer was reduced to 0.8 µm to shorten the transition time. Especially, the density for the charge sheet is designed to be about 3.5 x 10 12 /cm 2 to maintain the high electric field in the multiplication region. The back illumination structure was intended to lower the device capacitance. In addition, it is designed to compensate for the optical loss through the re-absorption of the reflected light beams at p- metal contact. For a shaped p-junction, a thin layer of three thousand angstrom is etched away at the central region of a main p-junction as shown in Figure 1. The shaped p-type junction finally results in a multiplication layer width (MLW) difference between main junction and floating guard ring region. Subsequently, only one sealed ampoule diffusion process was followed to construct an abrupt p-n junction and floating guard ring in APD. Therefore, the MLW was controlled by a Zn diffusion depth, ~ 2.9 µm in this work, and a recess etching thickness of the central region. After Zn diffusion, we designed the distance between main junction and floating guard ring to be 2.5 µm. In addition, three 0.05 µm-thick InGaAsP grading layers were used and P-metallization was achieved by alloyed Ti/Pt/Au contact. In order to form an abrupt p-n junction of the APDs, we used the sealed ampoule diffusion process, which is generally accepted as the preferred technique. Then, the multiplication layer thickness was controlled by a Zn diffusion depth and reactive ion etching depth of the central region. To reduce the number of diffusion, and to efficiently control the MLW (multiplication layer width) the recess etching was used. This structure has an advantage for ease of fabrication and for maintaining the reliability of the devices. The backside aperture was coated with anti-reflecting SiN x, to eliminate the reflection from the air-inp interface. The outside region of the aperture was capped with Cr/Au for n-electrode contact by photolithographic lift off, which also protects the stray light illumination. To achieve a high gain bandwidth product, a thin multiplication layer width (MLW) is essential to reduce the avalanche build up time 7,10. It also has many other good effects on high speed operations 11. Therefore, we must carefully control the diffusion depth to the range of sub micrometers. To reduce the prebreakdown at the curved junction region, we designed a shaped main junction and one floating guard ring, as shown in Fig. 1. Proc. of SPIE Vol. 4999 131

Current (A) 1x10-3 1x10-5 1x10-7 1x10-9 1x10-11 0 10 20 30 Voltage (V) Figure 2. Measured dark and photocurrent as a function of reverse bias for a typical 20 µm diameter of APD. Figure 2 shows the dark current and the 1.55 µm light illuminated photocurrent as a function of reverse bias for a typical 20 µmdiameter active area APD device as measured by HP 4145. The dark current has some noise currents induced by the measurement system. However, all the measured dark currents remain below several na s at the operating voltage, which are low enough to operate as a photodetector. The punch-through voltages (V p ) are between 11 V to 20 V according to the diffusion depth variation, i.e., MLW variation. However, the measured breakdown voltages are in the range of 25 V ~ 29V, which are not significantly varied as compared to the V p variation. The breakdown voltages are defined as the voltage where the photocurrents exceed 100 µa. As mentioned earlier, the speed of the APD depends mainly on the MLW. Thus the widths of the designed multiplication layer are expected to be between 0.25 ~ 0.35 µm considering the density of charge sheet and InGaAs absorption layer thickness. Previously reported results 7 suggest that there exists a minimum breakdown voltage as MLW increases for a given charge sheet density. At the minimum MLW, the transition time is shortened, and then the best performances and the best gain and bandwidth products are expected under the given epitaxial structures. Figure 3. (a) Measurement Breakdown voltage versus punchthrough voltages for all the three types. 132 Proc. of SPIE Vol. 4999

Figure 3. (b) Measurement Breakdown voltage versus punchthrough voltages for type I APD. The relation between punch-through voltage and breakdown voltage of APDs, fabricated on the same epitaxial wafer, are shown in figure 3 (a), (b). In figure 3(a), the wide spread of punch through voltage are shown, which are due to the fabrication (manufacturing) parameter variation and structure variation. Through the variations, three types of APDs are produced. The structure for type I APD is shown in figure 1, which has shaped main p-junction and one shallow guard ring compared to main p-junction. The structure for type II APD are fabricated with no recess etching and has one shallow guard ring. And, the structure for type III APD are also made with no recess etching but the depth of guard ring is deeper than that of main p-junction. All the three types of APD shows good avalanche behavior as judged from photo and dark current-voltage characteristics. Figure 3(b) shows Capacitance (ff) 300 200 100 0 5 10 15 20 Voltage (V) Figure 4. Measured capacitance versus reverse bias voltages of APD with 20 µm-diameter light receiving area. breakdown voltages to punch through voltages for type I APD, and the solid line represents the best fitted 2nd order polynomial curve. Figure 3(a) The punch through voltage is a function of MLW, that is, the smaller V p corresponds to the thinner MLW. From the results of Fig. 3 (b) and from the previous arguments, it is confirmed that multiplication layer thickness of fabricated APD is in the range of designed value. The important parameter to be used as a high speed detector is the capacitance of a device. Figure 4 shows the measured capacitance of APD chip, which was designed to have 20 µm diameter light receiving area. The measured capacitance is Proc. of SPIE Vol. 4999 133

0.27 pf at zero bias, which is reduced below 0.13 pf after the punch through voltage, 17 V. At typical APD operating voltage, which is typically above 90 % of breakdown voltage, the capacitance is estimated to be around 100 ff. Considering the RC time constant alone, we can say that it is low enough for the APD chip to be operated at high speed up to 10 GHz.. 2. Two dimensional gain profiles and Reliability Experiments To investigate the gain suppression at the pheripheral region of the light receving area, we measured the two dimensional photocurent characteristics of APD. If a focused beam illluminates a tiny part of the device, the photocurrent originating from only the illuminated part of the device can be detected. The focused beam was scanned to x- and y-direction on the APD surface, and then the photocurrent according to the spatial distribution was obtained. 1.2-40 -20 0 20 40 Distance(µm) 1.0 0.8 0.6 0.4 0.2 0.0 20 40 60 80 Current(Arb. Unit) Distance( µm) Figure 5. Two-dimensional photocurrent characteristics according to the x-y position of a focused beam on APD. The abrupt decrease of photocurrent caused by no light illumination on the p-metal contact region. Figure 5 shows the two-dimensional photocurrent characteristics of an APD at 90 % breakdown voltage, where the APDs are usually operated at 90% of breakdown voltage in 10 Gbps operation. The number on the x-y plane represents the real distance between the device center and measured points. We used the same APD structure as used in gain and bandwith measurement, except the light illumination direction. Because a focused light can not come into the curved region of the main junction in the backside illumination type, as shown in Fig. 1, we made the same structure with front illumination type APD for two dimensional photocurent measurement. In the front illumination type APD, only a 2-µm metal ring on the main junction interrupts light illumination, on which the photocurrent suddenly decreased as shown in Fig. 5. The diameter of light receiving area is 25 um as shown in Fig. 5 and z- axis shows the photocurrent obtained by the focused light beam diameter of ~1 µm. We can observe that the photocurrent at the curved region of the main junction is clearly decreased. In other words, the gain is successfully supressed at outside region of the active area including the floating guard ring region, as the photocurrent is proportional to the gain in APD. These results suggest that the shaped main junction and one floating guard ring works well in this APD structure. Fig. 6 shows the temperature dependence of the dark current characteristics. As the temperature varies from the room temperature to 200 o C, the dark current variation was confirmed to be within 10 300 na under the 0.9 V B bias conditions. It is observed that there is no excessive dark current increase compared with the mesa structured APDs indicating that the dark current degradation in floating guard ring region is not significant with respect to the temperature variations. However, the value of dark current significantly increased over 2 µa at 250 o C indicating that this condition can be regarded as a relatively high stress condition. 134 Proc. of SPIE Vol. 4999

1E-4 Dark current at 0.9 V B 1E-5 Dark current [A] 1E-6 1E-7 1E-8 1E-9 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 1000/T [1000/K] Figure 6. Temperature dependence of the dark current at 0.9 V B. Fig. 7 depicts the percent of cumulative failures versus the lognormal projection of the device time-to-failure after accelerated life testing. Although the sample size is small, in case the data appears linear, which indicates that the failure mode is the wear-out type. Failures obey the lognormal distribution relatively well. Median lifetimes for the devices at 200 and 250 o C were estimated to be each time. The Arrhenius plot of median lifetimes as a function of Cumulative Failure [%] 99 95 80 70 60 50 40 30 20 10 5 1 reciprocal aging temperature is shown in Fig. 8. From this plot, the thermal activation energy of the device aging process is computed to be 0.95 ev. Using this activation energy level, the median APD lifetime under practical use conditions can be estimated to be 3.9 x 10 9 hour at room temperature, with a standard deviation of 100 1000 10000 651 hour. Lifetime [hour] Due to the Figure 7. Lognormal projection of time-to-failure versus percent of lognormal degradation cumulative failures for APDs after life testing at 200 and 250 o C. behavior of the APDs, the failure probability of each device as a function of time, using the average device lifetime (µ) and its standard deviation (σ) as 13 : P () t 250 deg.c 1 = σ 2π t 0 1 exp t ( ln t µ ) 2σ 200 deg.c 2 2 dt Proc. of SPIE Vol. 4999 135

Along with the lognormal plot, this expression provides a quantitative method of evaluating the likelihood of failure for a given device as a function of its age. 1.E+08 1.E+07 Median Lifetime [hour] 1.E+06 1.E+05 1.E+04 1.E+03 Figure 8. Arrhenius plot of median APD lifetime as a function of reciprocal aging temperature. 1.E+02 1.5 2 2.5 3 3.5 1000/ T [1000/ K] 3. Frequency Response of APD chip The frequency response of APD was measured by varying the applied reverse bias voltages, which means different gain conditions. Figure 9 illustrates the gain and frequency response. APD was mounted on a metalized substrate with 150 µm holes for back illumination and ground-signal-ground probe with bias-t was used for RF output and DC bias input. Externally modulated tunable 1.55 µm-ld was used as a light source, and frequency responses were measured by using a RF spectrum analyzer. Prior to the measurement, all the components are measured and standardized to meet the purpose of our APD frequency measurement. The -3 db bandwidths as measured by varying reverse bias voltage from 20 V to 27.5 V are represented in figure 9. Finally, we achieved the gain and -3dB bandwidth product estimated as high as 80 GHz through the dotted line, which is due to the thin MLW and avalanche process in high electric field regime. -3dB Bandwidth(GHz) 100 10 1 1 10 100 Gain Factor Figure 9. Measured -3dB bandwidth of APD as a function of gain factor. 136 Proc. of SPIE Vol. 4999

III. SUMMARY We successfully demonstrated that the proposed APD operates well to the gain-bandwidth of over 80 GHz with good reliability. The APD fabrication process is rather very simple and the result of the two-dimensional photocurrent supports a highly reliable operation of the device. It is also believed that the APD with narrow MLW would have lower avalanche build up time and enhanced ionization coefficients ratios than a thick MLW InP/InGaAs APD. All the above results support that this APD structure that we have developed yielded device characteristics that are sufficiently good for practical optical receiver modules operating normally up to 10 Gb/s. References 1. L. E. Tarof, D. G. Knight, K. E. Fox, C. J. Miner, N. Puetz, and H. B. Kim, Planar InP/InGaAs avalanche photodetectors with partial charge sheet in device periphery, Appl. Phys. Lett, Vol. 57, No. 7, pp.670-672, Aug. 1990. 2. R. Kuchibhotla, J. C. Campbell, C. Tsai, W. T. Tsang, and F. S. Choa, Delta-doped SAGM avalanche photodiodes, IEEE Transactions on Electron Devices, Vol. 38, No. 12, pp.2705-2706, Dec. 1991. 3. I. Watanabe, S. Sugou, H. Ishikawa, T. Anan, K. Makita, M. Tsuji, and K. Taguchi, High-speed and low-darkcurrent flip-chip InAlAs/InAlGaAs quaternary well superlattice APDs with 120 GHz gain-bandwidth product, IEEE Photonics Technology Letters, Vol. 5, No. 6, June. 1993. 4. P. Yuan, O. Baklenov, H. Nie, A. L. Holmes, B.G. Streetman, Campbell, J.C, High-speed quantum-dot resonantcavity SACM avalanche photodiodes operating at 1.06 µm, 57 th Annual Device Research Conference Digest, 1999, pp. 202-203. 5. Y. Liu, S. R. Forrest, J. Hladky, M. J. Lange, G. H. Olsen, D. E. Ackley, A planar InP/InGaAs avalanche photodiode with floating guard ring and double diffused junction, IEEE Journal of lightwave technology, Vol. 10, No. 2, pp. 182-193, Feb. 1992. 6. Ackley D. E., Hladky J., Lange M.J., Mason S., Erickson G., Olsen G. H., Ban V. S., Liu Y., Forrest S. R. InGaAs/InP floating guard ring avalanche photodiodes fabricated by double diffusion. IEEE Photo. Tech. Lett., 1990; 2(8): 571-573 7. Kyung-Sook Hyun and Chan-Yong Park, Breakdown characteristics in InP/InGaAs avalanche photodiode with p- i-n multiplication layer structure, J. Appl. Phys, Vol. 81, No. 2, 15 Jan. 1997. 8. Chan-Yong Park, Kyung-Sook Hyun, Seung-Goo Kang, and Hong-Man Kim, Effect of multiplication layer width on breakdown voltage in InP/InGaAs avalanche photodiode, J. Appl. Phys, Vol. 67, No. 25, pp. 3789-3791, Dec. 1995. 9. P. Yuan, C. C. Hansing, K. A. Anselm, C. V. Lenox, H. Nie, A. L. Holmes, Jr., B. G. Streetman, and J. C. Campbell, Impact ionization characteristics of III-V semiconductors for an wide range of multiplication region thickness, IEEE journal of quantum electronics, Vol. 36, No. 2, Feb. 2000. 10. Anselm K. A., Nie H., Hu C., Lenox C., Yuan P., Kinsey G., Campbell J.C., Streetman B. G. Performance of Thin Separate Absorption, Charge, and Multiplication Avalanche Photodiodes. IEEE J. of Quantum Electron.1998; 34(3): 482-490 11. Hayat M. M, Kwon O-H, Pan Y., Sotirelis P., Campbell J. C., Saleh E.A., Teich M. C. Gain-Bandwidth Characteristics of Thin Avalanche Photodiodes, IEEE Trans. on Electron Devices 2002;49(5):770-781. 12. Jung J., Kwon Y.H., Hyun K.S., Yun I. Reliability of Planar InP/InGaAs Avalanche Photodiodes with Recess Etching. IEEE Photon. Tech. Lett. 2002;14(8):1160-1162. 13. F. Nash, Estimating Device Reliability, Norwell, MA: Kluwer, 1993 Proc. of SPIE Vol. 4999 137